Documente Academic
Documente Profesional
Documente Cultură
PIEZOELECTRIC ACTUATOR
2015/2016-I
Signature : ………………………………………………
Name of Supervisor : Ir. Dr Mohd Ridzuan Ahmad
Date : ........................................................................
MODELLING AND CONTROL OF HYSTERESIS AND VIBRATION IN
PIEZOELECTRIC ACTUATOR
DECEMBER 2015
ii
I declare that this project report entitled “Modelling and Control of Hysteresis and
Vibration in Piezoelectric Actuator” is the result of my own research except as cited
in the references. The project report has not been accepted for any degree and is not
concurrently submitted in candidature of any other degree.
Signature : ………………………………..
DEDICATION
ACKNOWLEDGEMENT
All praise and thanks is to Allah, The most beneficent and The most merciful!
I would like to use this medium to express my sincere gratitude to my project
supervisor, in person of Ir. Dr. Mohd Ridzuan Bin Ahmad for his relentless
assistance and guidance. My sincere appreciation goes to Auwal Shehu Tijjani, Md.
Habibur Rahman and Bashir Bala Muhammad for their immeasurable assistance.
Without their help, the success of this work would have not been achieved. May
Allah reward them in abundance. Amin!
I am also grateful to the Kano State government for giving us the opportunity
to further our studies. May Allah give us the strength and courage to help others to
the best of our ability too. Amin!
ABSTRACT
ABSTRAK
TABLE OF CONTENTS
DECLARATION ii
DEDICATION iii
ACKNOWLEDGEMENTS iv
ABSTRACT v
ABSTRAK vi
LIST OF TABLES x
LIST OF FIGURES xii
LIST OF ABBREVATIONS xv
LIST OF APPENDICES xvii
1 INTRODUCTION 1
1.1 Introduction 1
2 LITERATURE REVIEW 15
2.4 Summary 39
3 METHODOLOGY 42
function model 60
strategies 84
3.5 Summary 84
5 CONCLUSIONS 103
REFERENCES 105
Appendices A-E 113-123
x
LIST OF TABLES
LIST OF FIGURES
LIST OF ABBREVATIONS
2D - Two Directional
3D - Three Directional
BC - Boundary Condition
D - Electric Displacement
DOF - Degree Of Freedom
E - Electric Field
FE - Finite Element
FEA - Finite Element Analysis
FEM - Finite Element Model
GND - Voltage Ground
GUI - Graphical User Interface
IAE - Integral Absolute Error
Kd - Derivative gain
Ki - Integral gain
Kp - Proportional gain
LQR - Linear Quadratic Regulator
MCU - Microcontroller
MLA - Multilayer Actuator
PEA - Piezoelectric Actuator
PES - Piezoelectric Sensor
PID - Proportional Integral Derivative
PV - Process Variable
xvii
LIST OF APPENDICES
Model 114
INTRODUCTION
1.1 Introduction
The term piezoelectricity is derived from the root word “piezein” meaning
“pressure”. Hence, piezoelectricity originally means “pressure electricity” [1,2]. It is
a well known process that was first discovered by the Curie brothers in the year
1880 [3]. Their observation was that when pressure was applied to some materials,
an electrical potential (or charge) was developed. It was discovered later that when
an electrical potential is applied to these materials motion (e.g. contraction or
elongation) was observed [2,4-6]. To mention only few, materials that exhibit
piezoelectric property include ceramic materials such as lead zirconate titanate
(PZT), barium titanate (BaTiO2) [2,6] and other monocrystalline materials like
quartz, tourmaline, Rochelle salt, etc. Ceramic piezoelectric materials offer higher
piezoelectric effect when as compared to that exhibited by monocrystalline materials
[7]. Among the several piezoelectric materials that we have, the most commonly
used material is PZT [2,7,8].
Figure 1.2 shows the open loop and closed loop input-output behaviour of
piezoelectric actuator. The open loop behaviour is as a result of hysteresis effect.
Vibration is a repetitive motion that occurs after a time interval with respect
to a point of reference. Vibration theory deals with the study of vibratory or
oscillatory bodies and the forces that are associated with them. The elementary parts
of a vibrating body are [30]:
Potential energy storage means (spring or elasticity)
Kinetic energy storage means (inertia or mass)
A gradual energy damping means (damper)
Voltage, V
Figure 1.2 Open loop and closed loop behaviour of piezoelectric actuator
6
Finite Element Analysis (FEA) is a numerical method that is used for solving
multi-physics problems [31]. The approach involves dividing the continuous
geometry of the system into finite number of simpler units that are called elements.
These elements are connected together at nodal points. Unlike in analytical method
where the entire system solution is solved in one operation, FEA involves defining
the equations for every element and these equations are solved simultaneously to
obtain a complete system solution [32]. This is to say that FEA is an approximation
method.
The continuous system is discretized into smaller finite units called elements.
The smaller units can be solved more easily than the continuous whole system by
solving the partial differential equations for the elements. The solutions for the
equations representing the whole system will give us the approximate solution for the
continuous system. The accuracy of the result increases as the number of elements is
increased [31]. Figure 1.3 shows the general procedure involved in FEA.
Pre-processing
Post-processing Processing
Figure 1.3 General procedure for FEA
8
Figure 1.4 Polarization process. (a) original state (b) during polarization (c) after
polarization completed
9
Hysteresis
Bouc-
Wen
k m
u
b k
u
b
m b k u (1.1)
The major aims of the project are to improve the PEA system's positioning
accuracy and efficiency. In order to achieve these aims, we want to achieve the
following objectives:
Experimental works and finite element analysis (using ABAQUS 6.14) were
conducted for the purpose of collecting input and output transient data. System
identification techniques were employed for the purpose of establishing transfer
function models of the piezoelectric actuator. The FEA data transfer function
modelling was conducted in four steps; FE modelling and validation, data
preparation based on FEA, transfer function generation based on system
identification and model validation. From transient data collected from experiments,
vibration model and hysteresis model for the PEA were established based on
identified optimum parameters of Bouc-Wen hysteresis model.
Month MAR MAR MAR MAR APR APR APR APR MAY MAY MAY MAY JUN
Learning ABAQUS and Modelling
Piezoelectricity in ABAQUS
Research for Literature Review
Finite Element Modelling of Piezoelectric
Actuator Using ABAQUS
Project Chapter 1 Writing
Acquisition and Learning How to Use
Signal Processing for ABAQUS
Literature Review (Chapter 2) Writing
Finite Element Model Validation
Project Methodology Writing and
Presentation
Project Synopsis Submission to Faculty
(Form RP1-1)
Submission of Seminar Material to
Faculty
Seminar & Form RP1-2 Submission
Project Report & Form RP1-3 Submission
to Supervisor
Table 1.2: Phase 2 activities Gantt chart
Month SEP SEP SEP SEP OCT OCT OCT OCT NOV NOV NOV NOV DEC
Project proposal
Controllers Design
LITERATURE REVIEW
At this point, literature review is presented. The literature review divided into
three phases. In the first phase, a review on theoretical background necessary for well
established understanding of the concept of piezoelectricity is presented. In the
second phase, a review on past work is presented. This includes a review on
piezoelectric actuator modelling based on finite element analysis, hysteresis
modelling, hysteresis compensation and residual vibration control are presented. Last
phase of the review work presents a summary of past related works.
regarding how the transducers behave when they are under electrical and mechanical
loads in addition to information about their strength data. Y. Bernard [4] and B. J.
Sung et al. [36] reported that the linear equations of piezoelectricity when
considering the strain and the electrical field as independent variables are given by
Eq. 2.1 and Eq. 2.2. These equations are in agreement with the ANSI/IEEE 176
standard of piezoelectricity [37]:
S i sij T j d ji E j
E t
(2.1)
D j d ij Ti Tjk Ek (2.2)
According to [34], Eq. 2.1 and Eq. 2.2 are referred to as the strain formulation
while Jarolslaw Latalski [35] reported that the sets of Eq. 2.1 and Eq. 2.2 are known
as the 'displacement-based' formulation. The author reported that the equations are
the standard equations for describing piezoelectricity and added that the equations
are based on the assumption that the sum of mechanical-induced strain and
electrically-caused strain are what that produce the total strain in the transducer.
Sirohi and Inderjit in [38] reported, the piezoelectric stress formulation equations are
given by Eq. 2.3 and Eq. 2.4 while Jarolslaw Latalski [35] regarded these equations
as the 'force-based' formulation. Author in [26] stated that Eq. 2.3 represents the
inverse piezoelectric effect while Eq. 2.4 represents direct piezoelectricity and that
the two sets of Eq. 2.3 and Eq. 2.4 are referred to as the 'force-based formulation'.
Where dij is a 3x6 strain coefficients matrix (m/Volt or Columb/N), Sij is a 6x6
The strain coefficients matrix has only 5 non-zero elements when calculated in
the frame with third direction as the poling direction. The element d33 is the
coefficient that is responsible for longitudinal displacement, while coefficient d31 =
d32 account for transverse displacements and coefficient d15 = d24 produce shear
strain only when the applied potential is orthogonal to poling axis-3. The two
transverse displacement occur simultaneously [4,38]. The coefficients d31 and d32
have negative values while d33 coefficient has positive value. We can see that, when
d33 produces extension motion in axis-3, the two coefficients d31 and d32 produce
contraction in axis-1 and axis-2 respectively and vice versa. Clinton Y. K. Chee et al.
[34] reported that, provided that the elastic compliance matrix is invertible, its
inverse will give elastic stiffness and vice versa. Expressing equations Eq. 2.1 and
Eq. 2.2 in 3-D systems gives Eq. 2.5. The authors in [35] further reported that, for
those materials that exhibit 4 mm crystalline class or 6 mm crystalline class (e.g.
PZT, barium titanate-BaTiO3), Eq. 2.5 reported by Clinton Y. K. Chee et al. in [34]
is well simplified to Eq. 2.6. The simplification has to do with the symmetry of
electric, electro-mechanical and elastic properties.
18
the coupling matrix) dictates the design of sensing and actuation devices. Eq. 2.7
describes actuator mode while Eq. 2.8 describes sensor mode [35,39].
Actuator mode:
Sensor mode:
S1 T1
S T
2 0 0 0 0 d15 0 2 11 0 0 E1
S 3 T3
0 0 0 d15 0 0 0 22 0 E 2 (2.8)
S 4 d T
0 4 0 0 33 E3
S 5 31
d 31 d 33 0 0
T5
S 6 T6
For the case of actuator, the applied stress in this case is zero. Hence, Eq. 2.1
reduces to Eq. 2.9 as reported by [5]. The matrix form given by Eq. 2.7 reduces to
the form given by Eq. 2.10.
S i d ji E j
t
(2.9)
20
0 0 d 31
0 0 d 32
0 E1
0 d 33
Si E 2 (2.10)
0 d 24 0
d 0 0
E3
15
0 0
0
The relationship between the applied electric field and the produced strain is
linear as stated by Eq. 2.9 at lower electric fields [5]. This theoretical linearity
between the actuator driving voltage and the output deformation can be observed in
the work done by Xiaoning Jiang et al. [40]. Figure 2.1 is an input voltage against
actuator displacement plot comparing theoretical data and experimental data curves
as reported by [5]. At higher electric fields, hysteresis problem occurs due to
polarization reorientation when the electric field strength is increased. According to
Robert K. Lenzen et. al. [5], hysteresis are mostly noticed when the applied electric
field exceeds about 100 V/mm.
As it has been pointed out earlier under this section, the strain constants, dij
(m/Volt) are responsible for producing motion in their respective directions. They
define the mechanical motion production capability of the piezoelectric material.
These constants provide us with the picture of how much the material will
extend/contract in the jth direction given an applied potential in ith direction [5]. The
commercially available active piezoelectric devices uses the d33-effect (meaning the
devices produces mechanical deformation in the direction of the applied voltage or
poling direction) or the d31-effect (meaning that the device produces deformation in
direction perpendicular to poling direction [35].
However, Y. Bernard, J.-L et al. [4] reported that, in order for some tens of
nanometre displacements to be produced, the applied potential must be high enough
(hundreds to thousands of volts). The challenge faced however is that, the material
has an upper voltage limit set forth by the disruptive field of the material. On the
other hand, when the applied voltage is below a lower setting of the material,
polarization reversals occur [4,8].
PEAs that are made from a single layer of piezoelectric material are referred
to as bulk actuators. Bulk-type actuators when sandwiched in-between two electrodes
can be used directly as PEAs. However, due to the limitations of bulk-type actuators
outlined in the preceding paragraph, manufacturers employ many technique to
achieve higher strokes without exceeding the material maximum voltage limit [4].
One of the techniques employed by manufacturers is multi-layer stacking. Multi-
layer stack technique is discussed below.
22
Figure 2.2 shows MLA where the electrodes are spaced h metres apart.
Assuming the number of layers along the length of the MLA is n and the applied
voltage is V, the electric field, E, between each layer is given by Eq. 2.11. Equation
2.9 described the strain in a single layer (bulk-type) actuator. Using equation Eq. 2.9,
the strain in each layer is then given by Eq. 2.12. Ignoring the electrodes thickness,
the total thickness of the actuator is n x h. The net strain produced by the actuator is
then given by Eq. 2.12 while the net displacement, Δ(n x d), given by Eq. 2.13 [41].
V
E (2.11)
h
h V
Si ndij (2.12)
h h
V
h ndij h d ij nV (2.13)
h
V
h3 nd33 h3 d 33nV (2.14)
h
23
Piezoelectric
Electrodes ceramic layers
Piezoel
Electr
ectric ceramic
Electrodes
ode pins
h
layers
h Electr
V ode pins
Equation 2.13 highlights to us that the MLA displacement along the j-th
direction is proportional to the number of stacked layers, n, the applied potential in
the i-th direction and the dij coefficient of the piezoelectric material [41]. For a
material poled in the 3-direction, having voltage, V, applied across the poling
direction, the resultant stroke (or displacement) was given by Eq. 2.14.
polarization being the same while electric field opposite [42]. One layer contracts as
the other extends and the net effect is a bending motion.
As reported by [42], the formula for the bimorph bean bending motion, u, in a
place, x, assuming linearity in the variation of electric field through the beam
thickness, is given by Eq. 2.15,
e31U 2
u x (2.15)
Eh2
are presented. Some researchers used electrical equivalent circuit to model the PEA,
some model the PEA as a spring-mass-damper system while others uses hysteresis
and creep to model the PEA. Finite element method/analysis (FEM/FEA) technique
has been utilized by many authors to model, simulate and analyse the PEAs. In this
research work, we categorise the various methods into three categories:
Control systems design that employs PEAs and piezoelectric sensors (PESs)
needs an accurate knowledge about the relationship between inputs and outputs (or
transfer function, TF,) of the system. From numerical approach, these relationships
are not easy to be determined [43].
The authors in the previous paragraph lamented that the frequency domain
modelling involves four steps, these are:
Complicated
Structure
Mode Analysis
Yes
Time- Frequency-
domain Transient Analysis Frequency Analysis domain
excitation excitation
Time-domain Frequency-domain
Response Data Response Data
Time-domain Frequency-domain
Parameter Parameter
Identification Identification
Modify Modify
TF
Transfer Function Transfer Function TF
model modelling modelling model
Frequency- Time-domain
domain validation validation
No Is TF model Is TF model No
accurate? accurate?
Yes Yes
Transfer Function
model
The flowchart shown in Figure 2.4 consists of two sections. The upper section
is FEA while the lower section is parameter identification section. FE model of the
system is established in the FEA section and validation of the FE model is done
mathematically in order to investigate the model accuracy. In the lower section,
parameter identification techniques are used to come up with the TF model based on
response data obtained previously during FEA.
In the next coming sections, some FEM based works by some researchers are
being itemized.
M. H. Rahma et. al. [44] modelled the piezoelectric actuator in d33-mode but
with direction-2 (or y-axis) as both the polarization and excitation axis. This is also
clearly visible based on the PZT material properties that they use. With regards to the
relationship between the input and output, a linear relationship was observed.
Quick
mount PEA
(a) (b)
Figure 2.6 (a)Cantilever beam with piezoelectric actuator model (b)quick mount
PEA
The major aim Hoffman et. al. [45] performed simulation was to demonstrate
the effect of imperfect piezo-actuator/beam coupling imperfection on static beam
deflection and shear stresses.
31
Staworko and Tadeusz et. al. [3] reported that piezoelectric transducer model
can be obtained from phenomena that was first observed by the Curie brothers (i.e.
potential difference is proportional to strain of a piezoelectric material,
U kx (2.16)
The authors in [3] made it clear that, the net some of the received and
delivered electrical and mechanical energy is zero (see Eq. 2.17), hence Eq. 2.17
gives Eq. 2.18. Considering Eq. 2.16, we obtained Eq. 2.19,
Uq Fx 0 (2.17)
32
(Uq)
F (2.18)
x
F kq (2.19)
For harmonic excitations Eq. 2.16 and Eq. 2.19 become Eq. 2.20 and Eq. 2.21
respectively.
k
U V (2.20)
j
k
F I (2.21)
j
U Z eo Z em I
F Z Z m o V
(2.22)
em
Figure 2.8 shows the electrical equivalent circuit model for piezoelectric
transducer [3].
33
I. Kelvin-Voigt Model
Considering a single element and a sinusoidal vibration given by Eq. 2.23, the
restoring force is given by Eq. 2.24 [6].
y a sin(t ) (2.23)
b 2
f ky by ka sin(t ) ba cos(t ) ky a y2 (2.24)
a
F i
K
y i 1 c
n
1 e i
b
(2.25)
Ki
y V
y sin(V ). p (2.26)
d V .b b
exp( (V B))
d (V ) (2.27)
(1 exp( (V B)))2
According to Narges Miri et al. [6], this model is a subclass of the Preisach
model that is originally designed for ferromagnetic materials, but easier to implement
than the Preisach model. The model according to the above authors in [6] is a
36
y(i), H(i), i and ɷ are displacement, backlash operator, backlash operator width
and slope of the backlash in which:
where V(i) and r represent the voltage and the width of the input respectively and
H(i) is known as the play operator where the initial condition is given by Eq. 2.30.
One of the most popular operator-based hysteresis model. The model captures
the hysteresis behaviour in nonlinear systems. The general expression for Preisach
model uses a double integrator in continuous form as given by Eq. 2.31 [63],
where X(t) represents actuator displacement and μ(α,β) is a weighting function that is
selected based on experiment and experience [63].
The model was identified by Bouc and modified by wen. Consider a 1-DOF
actuator with hysteresis nonlinearity. Its motion equation is given by Eq. 2.32,
where the parameters m, b, k, u, , de and h(t) represent the PEA mass, viscous
damping coefficient, elastic stiffness constant, driving voltage, PEA tip
displacement, piezoelectric strain coefficient and hysteresis displacement of the PEA
respectively.
n1 n
h u u u h u h , h(t0 ) h0 (2.34)
h(t ) u u u u h (2.35)
38
where α, β and γ are the parameters that define the shape, orientation and magnitude
of the hysteresis loop.
Physics based models are built on corresponding physical effect based on first
principle [60]. They are developed in order to predict piezoelectric actuator
displacement profile from input voltage. Narges Miri et al. [6,64] reported that the
physics based models give better definition of the relationship between the parts of
the system dynamics. The major limitation with regards to using physics based
models is parameters estimation. Rather than optimization techniques, Ad-hoc
techniques are usually employed for their parameters identification. Commonly used
parameter identification techniques for phenomenological based models are the
particle swarm optimization (PSO) [14,15,50,51,57,65], nonlinear least square [61],
evolutionary algorithms based [48,49,59,66,67] etc. However, evolutionary
algorithms have complex algorithm that require high cost of computation and have
slow solution convergence compared to other techniques like the simple direct search
[68]. The PSO does not consider the gradient for the problem to be optimized hence
an optimal solution may not be obtained.
2.4 Summary
METHODOLOGY
Start
Model Requires
Parameters
Data collection Identification?
Experiments
Optimized Control
System Design
Image Processing
System ID Acceptable
Response?
Figure 3.2 shows the platform for conducting the experiments. The prime
objective of the experiments are to capture hysteresis nonlinearity and the vibration
dynamics of the PEA system by capturing input and output data. A resin coated type
multilayer stack type PEA manufactured by NEC Tokin with item number
AE0505D16DF is the PEA system under study. As stated in the device datasheet [77],
the actuator maximum driving voltage is 150V DC with a corresponding maximum
output displacement of 17.4±2.0 μm. Hence the actuator's maximum displacement at
150 V ranges from 15.4 μm to 19.4 0 μm. Equipment and devices used for conducting
the experiments include: Olympus IX73-ICS inverted type microscope, a desktop
computer hooked to Olympus IX73-ICS, DC power supply unit, a laptop computer,
stack-type piezoelectric actuator, Arduino Uno R3 microcontroller module and signal
switching circuit. DC power supply unit was used to generate the required voltage that
was used for driving the PEA, while Arduino microcontroller based switching circuit
connected to a laptop computer was used for generating different signal wave forms.
The schematic of the microcontroller based switching circuit is given by Figure 3.3.
Signal from the microcontroller modulates the DC voltage across the collector-emitter
terminals of the transistor Q1 hence, based on current transistor configuration, the
signal reaching the PEA has the same frequency as the signal from the
microcontroller, but swings from 0 V to nearly the amplitude of the DC power source.
Software programs on a laptop computer were programmed onto the microcontroller
in order to generate the input signals' frequencies and wave forms of interest. The
PEA system was tested using different signal waveforms. Figure 3.4 shows a
triangular type input voltage that corresponds phase and unloading (decreasing
voltage) phase, while Figure 3.5 shows the corresponding change in tip position. The
PEA was also tested with a square wave type input signal vibrating at a frequency of
1Hz with a duty cycle of 50% as shown in Figure 3.6 while Figure 3.7 shows the
corresponding changes in tip position for the vibrating actuator at 1Hz. When an input
signal is applied to the PEA, change in the PEA's tip position (or displacement) were
45
captured using the IX73-ICS inverted microscope. A video was recorded at 25 frames
per second with the help of the inverted microscope manufacturer software known as
DigiAcquis.
140
Loading Phase
Unloading Phase
120
100
Voltage (V)
80
60
40
20
0
0 5 10 15 20 25 30 35 40 45 50
Time (s)
Figure 3.4 Triangular form for increasing and decreasing input voltage signal
15
Tip Displacement (um)
10
0
0 5 10 15 20 25 30 35 40 45 50
Time (s)
Figure 3.5 Output displacement for increasing and decreasing input voltage signal
47
70
60
50
Voltage (V) 40
30
20
10
0
0 50 100 150
Time (s)
5
Tip Displacement (um)
0
0 50 100 150
Time (s)
Figure 3.7 Output displacement for 1Hz squarewave input voltage signal
Free space
(a)
Actuator tip at 0V
Free space
(b)
Figure 3.8 Piezoelectric actuator tip displacement monitoring using inverted
microscope. (a) Tip at 125 V input. (b) Tip at 0 V input
exhibits hysteresis nonlinearity. The actuator has two distinct displacement profile for
increasing voltage (loading case) and decreasing voltage (unloading case).
Table 3.1: Input and output data collected from hysteresis experiment
Reading Increasing Increasing Decreasing Decreasing
Serial Voltage Displacement Voltage (V) Displacement
Number (V) (μm) (μm)
1 0 0.000 125 14.664
2 5 0.690 120 14.492
3 10 1.208 115 14.147
4 15 1.725 110 13.974
5 20 2.243 105 13.802
6 25 2.933 100 13.457
7 30 3.451 95 13.112
8 35 4.141 90 12.939
9 40 4.831 85 12.594
10 45 5.348 80 12.422
11 50 6.038 75 11.732
12 55 6.556 70 11.214
13 60 7.073 65 10.696
14 65 7.764 60 10.424
15 70 8.454 55 9.806
16 75 8.971 50 9.361
17 80 9.489 45 8.371
18 85 10.179 40 7.826
19 90 10.696 35 7.509
20 95 11.386 30 6.919
21 100 11.904 25 6.201
22 105 12.594 20 5.556
23 110 13.112 15 4.966
24 115 13.457 10 4.148
25 120 14.147 5 3.258
26 125 14.664 0 2.296
50
15
Loading Phase
Unloading Phase
10
Displacement (um)
0
0 20 40 60 80 100 120 140
Input Voltage (V)
At this point, the transfer function model of the actuator is obtained from the
rising and falling input data given by Table. 3.1. The MATLAB R2013a System
51
identification toolbox was used for the modelling process. The transient input and
output data given by Table 3.1 was imported into the MATLAB workspace window
for the purpose of creating time domain data object as shown in Figure 3.11. The
imported data was further processed by detrending to remove offset means. The next
step performed was to select the type of estimation. Transfer function was selected as
shown in Figure 3.12. A couple of models were generated as shown in Figure 3.13 by
trying different combination of zeroes and poles until a model that gives a better
estimate with regards to high percentage of fitness to validation data and lower MSE
and FPE errors. As given by Eq. 3.1, a fourth order model having a single zero was
found to give the best fitness with low MSE and FPE. The step response for the
model of Eq. 3.1 is shown in Figure 3. 14 while the time response criteria for the
model are given by Table 3.2.
3400S 62060
G S (3.1)
S 4 572.5S 310120S 2 61670S 539200
Amplitude (um)
Time (sec)
At this juncture the validity of the established given by Eq. 3.1 was
investigated. The model fits time domain validation data with 98.01% fitness and
MSE of 0.0081 and FPE of 0.0328. Based on this results, there is high percentage
fitness and very error levels hence, we can be rest assured that the established model
is valid. To further support this claim, residual analysis was employed to investigate
the model validity. The autocorrelation (Figure 3.15b) and cross correlation (Figure
3.15a) for the model are given by Figure 3.15. Based on 99.9% confidence interval
setting, we can see from Figure 3.15 that the model has passed validation tests based
on residual analysis. As shown in the given figure, it can be seen that majority of the
sampled data is within in the 99.9% confidence interval limit (given by the broken
lines) for autocorrelation and all the sampled data is within the confidence interval
limit for cross correlation examination.
54
Autocorrelation
0.5
-0.5
-20 -15 -10 -5 0 5 10 15 20
1.5
Cross correlation
1
0.5
-0.5
-1
-1.5
-20 -15 -10 -5 0 5 10 15 20
Samples
Figure 3.15 Residual analysis (a) auto correlation (b) cross correlation
The time response for the model were presented in Table 3.2. Based on these
criteria, we proceed by establishing a reduced second order model having similar
time response criteria. The general form for second order model is given by equation
3.2,
K pn2
G(s) (3.2)
s 2 2n s n2
where ωn is the natural frequency of the system, ζ is the damping factor of the
system and Kp is the system's final value. The two unknowns ζ and ωn were obtained
using Eq. 3.3 and Eq. 3.4 respectively.
55
ln(OS )
(3.3)
2 ln 2 (OS )
4
n (3.4)
Ts
ln(0.528)
Using Eq. 3.3, , 0.1992
2 ln 2 (0.528)
4
Using Eq. 3.4, n , n 8.0321 rad / s
2.5 0.1992
7.4192
Hence, from Eq. 3.2 we have: G(s) (3.5)
s 3.200s 64.5151
2
By comparing Eq. 3.5 with Eq. 3.6 the transfer function parameters for the
actuator were extracted and presented in Table 3.3.
1
G( s) (3.6)
ms 2 bs k
Model order reduction validity was investigated by comparing the response for
the fourth order model and the reduced second order model. As shown in Figure
3.16, the two models have similar characteristics.
56
0.18
Reduced Order Model
0.16 Original Model
0.14
0.12
Amplitude
0.1
0.08
0.06
0.04
0.02
0
0 1 2 3 4 5 6 7 8 9 10
Time (seconds)
Figure 3.16 Model order reduction validation based on time response criteria
The hysteresis model for the piezoelectric actuator was developed under this
section. Bouc-Wen model has been selected for the modelling the fact that it has few
parameters to be identified and its implementation in simulink is feasible. The
simulink block diagram shown in Figure 3.17 is based on Eq. 2.35 while the
complete hysteresis model shown in Figure 3.18 is based on Eq. 2.32. The transfer
57
function for the piezoelectric actuator has been established under section 3.2.1 while
the corresponding reduced order model was established under section 3.2.1.2. The
next step taken was Bouc-Wen model parameters identification based on Nelder-
Mead simplex search optimization.
H(u)
T. Fcn.
-
r 𝒅𝒆 +
K G(S)
N 1
A simplex, S is defined as a convex hull having N+1 vertices x j
j 1
in an
An iteration starts from an initial simplex. Kth iteration starts by ordering and
N 1
j1 such that
labeling the current vertices as x j
(k )
(k ) (k ) (k ) (k )
g1 g0 g3 ......... g N 1 (3.7)
(k ) (k )
The point x is referred as the best point, point x2 as next to the worst,
N 1
(k )
point x1 as the worst point.
59
The goal is to minimize the cost point. For this reason, worst point is
discarded and many 'better' trial points are generated. Function values are
computed at the generated points.
Next, a new simplex is constructed having N+1 vertices by utilizing rules that
favour minimization of the objective function.
At the 262nd iteration, the solution converged. The trajectories for the Bouc-
Wen parameters alpha (α), beta (β) and gamma (γ) are shown in Figure 3.19a, Figure
3.19b and 3.19c respectively, while the trajectory for piezoelectric coefficient is
given by Figure 3.19d. The optimum model parameters values are tabulated in Table
3.4.
Alpha
alpha
0.05
0
0.02 (a)
Beta
beta
0.01
Parameter Values
0 (b)
0.05
Gamma
gamma
-0.05 (c)
0.4
Piezoelectric Coefficient
strain
0.2
0
0 50 100 150 200 250 300
Iterations
Iterations
(d)
Figure 3.19 Parameters trajectories (a) alpha (b) beta (c) gamma (d) piezoelectric
coefficient.
At this juncture the transfer function model for the vibrating piezoelectric
actuator was developed. From the experimental procedure reported under section 3.2
of this work, input and output data for the vibrating PEA was collected. The
squarewave type input signal and the corresponding output displacement have been
shown in Figure 3.6 and Figure 3.7 respectively under section 3.2. Input and output
data obtained using image processing techniques discussed earlier under section 3.2.
At a sampling rate of 4Hz a total of 400 input and output data sets were collected for
the purpose of establishing the vibrating PEA's transfer function based on system
61
identification technique. Table 3.5 is a sample of the input and output data for the
vibrating PEA. Complete data is attached at the appendix of the report. Using the
same system identification procedure discussed in section 3.2.1, the transfer function
for the vibrating PEA given by Eq. 3.8 was obtained. Step response for the vibrating
PEA is given by Figure 3.20 while time response criteria for the system by Table 3.6.
Table 3.5: Samples of input and output data for vibrating PEA
S/N Voltage Displacement S/N Voltage Displacement
(V) (um) (V) (um)
1 0.063 0.000 31 0.063 6.543
2 0.063 0.000 32 0.063 6.543
3 63.050 6.543 33 63.050 0.000
4 63.050 6.543 34 63.050 0.000
5 0.063 0.000 35 0.063 6.543
6 0.063 0.000 36 0.063 6.543
7 63.050 6.543 37 63.050 0.000
8 63.050 6.543 38 63.050 0.000
9 0.063 0.000 39 0.063 6.666
10 0.063 0.000 40 0.063 6.666
11 63.050 6.666 41 63.050 0.000
12 63.050 6.666 42 63.050 0.000
13 0.063 0.123 43 0.063 6.666
14 0.063 0.123 44 0.063 6.666
15 63.050 6.543 45 63.050 0.123
16 63.050 6.543 46 63.050 0.123
17 0.063 0.000 47 0.063 6.543
18 0.063 0.000 48 0.063 6.543
19 63.050 6.420 49 63.050 0.123
20 63.050 6.420 50 63.050 0.123
21 0.063 0.000 51 0.063 6.666
22 0.063 0.000 52 0.063 6.666
23 63.050 6.543 53 63.050 0.000
24 63.050 6.543 54 63.050 0.000
25 0.063 0.000 55 0.063 6.666
26 0.063 0.000 56 0.063 6.666
27 63.050 6.666 57 63.050 0.123
28 63.050 6.666 58 63.050 0.123
29 0.063 0.000 59 0.063 6.666
30 0.063 0.000 60 0.063 6.666
62
-0.005
-0.01
-0.015
Amplitude (um)
-0.02
-0.025
-0.03
-0.035
-0.04
-0.045
0 5 10 15
Time (sec)
Time (sec)
0.8498
G( s) (3.8)
s 0.8194s 34.8
2
Figure 3.21 shows residual analysis results for the vibrating PEA. System
identification results shows that the established model fits validation data by 96.32%
63
1
Autocorrelation
0.5
-0.5
-1
-20 -15 -10 -5 0 5 10 15 20
(a)
1
Cross correlation
0.5
-0.5
-1
-20 -15 -10 -5 0 5 10 15 20
Samples
(b)
Figure 3.21 Vibrating PEA residual analysis (a) autocorrelation (b) cross correlation
fitness level with an MSE of 0.01468 and FPE of 0.01559. Based on the degree of
fitness and very small error levels, the model has passed validation test. Figure 3.21
shows residual analysis validation where Figure 3.21a showed autocorrelation of the
residuals for input voltage while Figure 3.21b showed cross correlation for input
voltage and output displacement. It can be seen all sampled data is within the 99.9%
confidence boundary. Hence, we can conclude that the model has passed validation
test from residual analysis perspective.
64
II. Step 2 (finite element meshing): in this step the finite element mesh is
applied to the 3D part. The part is meshed with element type C3D8E
(an 8-node linear piezoelectric brick) type of mesh with linear
geometric order, under standard element library. The meshed part
65
(a) (b)
contains a total number of 800 elements (of type C3D8E) and a total
number of 1089 nodes. Figure 3.23 showed the meshed part with
approximate global size of 0.5mm.
properly under the property module. These parameters that are required
are: elastic compliance (or stiffness) constants, any of piezoelectric
strain or stress coefficients, dielectric permittivity constants (or
permittivity) and material density. These properties depends on the type
of the piezoelectric material being used. For the purpose of this work, a
PZT-5H piezoelectric material is selected because it has the largest
value of longitudinal strain coefficient (d33-coefficient). The PZT
materials that were used for the material property definition under the
material property module are summarized in Table 3.7. The material
properties are defined in ABAQUS as a z-axis polarized material (or
d33-mode).
Table 3.7: Summarized material properties for a z-poled PZT-5H ceramic material
Parameter (unit) Value
IV. Step 4 (Section creation and assignment): The next step carried out was
section property definition to the part model. This was done using the
section module under the model tree. Figure 3.24 shows the graphical
user interface (GUI) for the section module. The section for the
piezoelectric layer was defined as a solid homogenous type section.
This is because the PZT-5H material is solid in nature and is also a
homogenous material. The z-poled material property defined under the
previous step before this were linked to the section that is created under
the current step. Section assignment to the 3-dimensional part model
was then performed by accessing the section assignment sub module
located under the part module tree. The GUI window for the section
assignment is illustrated in Figure 3.25. The section to be assigned was
selected using the combo box (drop down menu) that is next to the label
that is named 'section' under the section group on the 'edit section' GUI.
68
VII. Step 7 (analysis step creation): the step module was used to create an
analysis step. By default, ABAQUS creates a step names initial and
user can create other additional steps. The step is necessary for us to be
able to capture changes in the boundary conditions (which we will
create later). The step was named 'VOLTAGE_APPLIED' and is of the
type general static as shown in Figure 3.27a. The time period for the
analysis step was limited to two seconds with a fixed increment of 0.1
second which is sufficient for enough data to be written during the
71
(a) (b)
Figure 3.27 (a) Step creation user interface (b) Output variables request
VIII. Step 8 (Defining Boundary Condition, BC): the first boundary condition
created was an encastre type boundary condition (BC). This was
applied to the to the base (reference) of the piezoelectric layer as it can
be seen in Figure 3.28. This boundary condition holds the actuator base
by disallowing rotational and translational displacement in all
directions. This was done through the Boundary Conditions (BCs)
72
module and was created under the default step initial. The BC was set to
propagated through all other analysis steps. The next BC applied to the
piezoelectric layer was electric potential ground (GND). This was applied
to the face coloured red as it can be seen in Figure 3.29. This BC provides
electric potential ground for electric potential differences that we will apply
in the future to the piezoelectric layer. This BC was created under the main
analysis step titled 'VOLTAGE_APPLIED'.
The next step carried out was electric potential BC definition. The BC
face was picked from the viewport. The face this BC was applied is the
same face that encastre BC was applied earlier. An electric potential of
150V was applied as shown in Figure 3.30. The BC was defined under
'VOLTAGE_APPLIED' step.
X. Step 10 (job creation): having set and applied all the necessary steps
and conditions required in developing FE model of the piezoelectric
actuator, the next step performed was creating a job. Using the job
manager GUI the model was submitted for processing. When the job
was completed, results were available for simulation and other post-
processing steps.
As it was given by Eq. 2.9 and Eq. 2.10, the displacement in both the
longitudinal and transverse directions are dependent on the magnitude of the applied
voltage (or electric field) and the respective strain coefficient in the direction.
76
l3 V
Using Eq. (2.9): S i d ji E j Strain, S 3 d 33 3
t
l3 l3
w2 V
Using Eq. (2.9): S i d ji E j Strain, S 2 d 32 3
t
w2 l3
w2 5
2w2 (displacement2 ) d 32 V3 2.74 10 10 150 5.1375 10 8 m
l3 4
5.1375 108
Hence, w2 (displacement2 ) 2.5688 108 m
2
Based on finite element analysis input and output data, we proceed with
transfer function model generation using the same system identification technique
initially discussed under section 3.2.1. At a sampling interval of 0.1 seconds a total
number of 41 data sets were collected for the identification. The sample values of the
data sets were presented in Table 3.8. However, completed data sets have been
included in the appendix-C section of this report. The transfer unction model with the
highest fitness level and lowest MSE and FPE was a fourth order model with four
zeroes given by Eq. 3.9.
0.5
Autocorrelation
-0.5
Cross correlation
0.5
-0.5
-1
-20 -15 -10 -5 0 5 10 15 20
Samples
(b)
Figure 3.33 FEA model residual analysis (a) autocorrelation (b) cross correlation
Amplitude (um)
Time (sec)
At this point the various control systems for hysteresis minimization and
residual vibration suppression were designed. The design task has been divided into
two branches, a branch each for hysteresis minimization and residual vibration
suppression.
The PEA transfer function model (Eq. 3.5) stability shows that the system has
poles at (1.6 j 7.9) . Since all the system poles are located on the left half of the s-
plane this shows that the system is stable.
3.2 64.52 1
A , B , C 0 7.42
1 0 0
81
PID control is the most common type of feedback control. More than 95% of
control loops in process control loops are of PID type [79]. In practice all PID
controllers are realized using microprocessor or microcontroller. This added features
like automatic gain tuning and gain scheduling [80]. The PID controller has
extensive functionality and applications, but due to the lack of proper understanding,
the full power of PID is not efficiently utilized[81]. PID has proportional, integral
and derivative actions. The functions performed by each action are summarized in
Table 3.10. The gains Kp, Ki and Kd determine how much of the proportional,
integral and derivative actions are used. Figure 3.35 shows a simplified parallel form
Figure 3.35 Simplified block diagram for parallel form PID control
82
PID. It basically explains what a PID controller does. The difference between the
setpoint (SP) and the process variable (PV) gives the error. The error gets multiplied
by one of the PID action or two or all the three PID actions depending on which
actions are turned on, and the result obtained gives the PID controller output (or
control effort) given by uc(t).
The PID controller algorithm is given by Eq. 3.10. The error, e(t) given by
Eq. 3.11 is the tracking error obtained as the difference between the reference
setpoint and the actual output [82]. The hysteresis nonlinearity problem here is
treated as the error that occurs in tracking the desired output displacement. Block
diagram for the hysteresis compensation based PID control scheme is given by
Figure 3.36 where the Bouc-Wen hysteresis model, H(u) stiffness constant, k and
transfer function model G(s) all have been established before this point. The tuning
parameters Kp, Ki and Kd for the PID controller given by Table 3.11 were obtained
using Nelder-Mead simplex search optimization technique. Initial guess for the
tuning parameters were established using MATLAB automatic tuning feature in
order to speed up the optimization process.
d
uc t K pe t Ki e t dt Kd e t (3.10)
dt
e t r (t ) (t ) (3.11)
83
H(u)
h T. Fcn.
uc -
𝒅𝒆 +
K G(S)
Bouc-Wen model
-
PID
+ r
Table 3.11: Optimum values for the PID controller tuning parameters for hysteresis
minimization.
Initial Guess Optimal Gains
Kp Ki Kd Kp Ki Kd
52.6280 70.6480 3.9813 50.1370 84.4050 3.9926
r (t ) (t ) 0 (3.12)
84
(t ) deu H (u ) (3.13)
Considering Eq. 3.13, if we consider only the linear term given by deu, we
can extract the value of u such that the desired displacement is, r is satisfied. By
doing this, Eq. 3.14 was obtained. The compensator is given by Eq. 3.14. The
compensator has the displacement of interest, r as the input command, a nonlinear
feedback, H(u) and output, u. Unlike most model inversion based hysteresis
minimization techniques, no complex calculations were needed to obtain an inverse
hysteresis model. The only required inversion the technique requires is that of de and
is strictly positive. Since de and H(u) have been identified earlier, no further
calculations are needed. For this reason, in terms of computational cost the
compensator implementation is simple.
1
de r
u H (u ) (3.14)
H(u) H(u) h
𝟏 u -
+ 𝒅𝒆 +
r 𝒅𝒆
+
G(S) K
At this point the various strategies that were employed for residual vibration
suppression are presented. LQR, ZV input shaper, pole placement and PID control
strategies were investigated for level of residual vibration suppression. The various
strategies are discussed in the next coming sub-sections.
The PEA transfer function (Eq. 3.8) stability shows that the system has poles
at (0.41 j5.88) . Since all the system poles are located on the left half of the s-
plane, it shows that the system is stable.
0.82 34.80 1
A , B , C 0 0.85
1 0 0
G o C CA respectively have full ranks. Hence the system is controllable and
T
observable.
86
x(t )
J T
Qx(t ) U (t )T RU (t ) dt (3.15)
The main objective of LQR control is to maintain state deviation close to zero
and keep the control minimal. If ones objective is to keep the state x(t) near zero,
then it is known as state regulator system while in the case where the objective is to
maintain the state near a desired value, it is known as tracking [84].
Figure 3.38 shows simulink diagram for the LQR control strategy. Tuning
matrices Q and R are given by Eq. 3.16, while Eq. 3.17 gives the control law
feedback gain, K and steady state error compensator, N obtained based on Eq. 3.16.
443 0
Q , R 1 (3.16)
0 335
87
Figure 3.38 LQR based vibration control structure with IAE measurement capability
For high precision positioning it is required that the residual vibration in the
positioning system should be zero. Zero vibration (ZV) input command shaping is an
open-loop technique for removing residual vibration in under-damped or undamped
systems [73].
tn
2 2
V (0 , ) e 0 C 0 , S 0 , (3.18)
89
where,
n t
C (0 , ) Ai e 0 i cos( w ti ) (3.19)
i 1 d
n t
S (0 , ) Ai e 0 i sin( w ti ) (3.20)
i 1 d
Ai and ti are the amplitudes and time locations of the impulse signals, where n
Ai 1 (3.21)
Ai 0, i 1, 2,........, n (3.22)
90
At this juncture, the task is to find the sequence of impulse amplitudes that
make Eq. 3.18 to zero at the same time the constraints defined by Eq. 3.21 and Eq.
3.22. For the problem in this work, there are four unknowns -impulses amplitudes (A1
and A2) and their corresponding time locations (t1 and t2). Without lost of generality,
the time location for the first amplitude (A1) can be set to zero [73] in order to avoid
unnecessary time delays, hence there are three more unknowns that need to be
determined. For Eq. 3.18 to be zero, both Eq. 3.19 and Eq. 3.20 are required to be
independently zero, because the two equations are squared in Eq. 3.18. Hence, the
two impulses must satisfy Eq. 3.23 and Eq. 3.24.
0t2
A1 A2e cos( w t2 ) 0 (3.23)
d
0t2
A2e sin( w t2 ) 0 (3.24)
d
n nTd
w t2 n , t2 , n 1, 2... (3.25)
d w 2
d
where Td is the damped period of vibration. In order for the vibration to be cancelled
in the shortest amount of time, the smallest value of t2 is chosen as
T
t2 d (3.26)
2
91
In this case, the amplitude constraint defined by Eq. 3.21 reduces to Eq. 3.27.
A1 A2 0 (3.27)
If we substitute Eq. 3.26 and Eq. 3.27 into Eq. 3.23, the result is given by Eq.
3.28. Reformatting Eq. 3.29 gives Eq. 3.29.
A1 (1 A1) exp( ) (3.28)
1 2
exp( )
1 2
A1 (3.29)
1 exp( )
1 2
By defining K exp( ) , the impulse signals can be summarized in matrix
1 2
form given by Eq. 3.30.
1 K
Ai
1 K 1 k (3.30)
ti 0 0.5T
d
From the vibration model earlier obtained and presented as Eq. 3.8, we can
extract the following:
92
0.8194
20 0.8194, 0.06945 (3.32)
20
0.2182
Hence, K exp( ) exp ( ), K 0.8036
1 2 0.9976
1 K
Therefore, A
1 1 K 0.5545
. Similarly, A
2 1 K 0.4456
Time location for A2 is t 0.5338s
2
0 1 2
Figure 3.41 ZV input shaper control structure with IAE measurement unit
The idea is to have the closed loop system matrix modified. The parameter K
is the feedback gain that ensures the system is forced to meet design specifications.
Modified closed loop system matrix, Ac is given by Eq. 3.34,
Ac ( A BK ) (3.34)
where A is the original system matrix and B is the system's output matrix.
c ( s) s ( A BK ) 0 (3.35)
In order to extract the feedback gain vector, K the coefficients of the terms in
Eq. 3.33 and those of the terms in Eq. 3.35 are compared directly.
The design requirements in this work are to have a closed-loop system with a
maximum overshoot of 2% and a settling time of 0.5 seconds. Using these
specifications, the task is to build a second order model with characteristics equation
given by Eq. 3.36.
The parameters and are calculated using Eq. 3.37 and Eq. 3.38 respectively,
0
1 2
O.S e (3.37)
4
Ts 2 (3.38)
0
By comparing Eq. 3.39 with Eq. 3.35, the gain matrix, K is given by Eq. 3.40. The
compensator, N is a pre-compensator that helps for zero steady state error to be
achieved. Its value is scaled using Eq. 3.40.
The PID control strategy has been has discussed under section 3.4.1.1. Under
this section, we present the optimum PID tuning parameter values (Kp, Ki and Kd).
The values for the PID tuning parameters given by Table 3.13 were obtained based
on Nelder-Mead simplex algorithm. The simulink implementation of the PID based
vibration control strategy is shown in Figure 3.42.
3.5 Summary
At this point the various results obtained are presented and discussed. The
results presented include simulation results from identified Bouc-Wen model, results
obtained from the various control strategies for hysteresis minimization and results
for residual vibration control based on the different control strategies reported in this
work. The degrees of hysteresis minimization and residual vibration control by the
different control systems are investigated based on integral absolute error.
Earlier, the optimum parameter for the Bouc-Wen models have been
identified and were presented in Table 3.4 under section 3.2.2.1. Figure 4.1 shows a
comparison between the real physical PEA system and simulated data based on the
established Bouc-Wen model. It can be seen that the simulated data is in agreement
98
16
Measured data
14
Simulated data
12
10
Displacement(um)
-2
0 20 40 60 80 100 120 140
Input Voltage (V)
Figure 4.1 A graphical comparison between measured data and simulated data
obtained from identified Bouc-Wen hysteresis model.
with data measured experimentally. Referring to Figure 4.1, the maximum hysteresis
gap is about 3 μm. This gap difference represents a 3 μm difference in recorded
output displacement between rising input voltage levels and decreasing voltage
levels. For most applications requiring high precise positioning, positioning error
may be up to about 3 μm. In order to quantify the magnitude of the error due to
hysteresis, IAE was employed. The IAE error for the uncontrolled hysteresis was
recorded as 67.73. This IAE value was set as the standard upon which the
performances of the various hysteresis minimization control techniques in this work
were evaluated.
99
16
12
Displacement (um)
10
-2
0 20 40 60 80 100 120 140
Voltage (V)
16
12
10
Displacement (um) 8
-2
0 20 40 60 80 100 120 140
Voltage (V)
0.1
PID Control Error
Inverse Control Error
0.05
-0.05
-0.1
-0.15
0 5 10 15 20 25 30 35 40 45 50
Time (s)
Figure 4.4 Reference tracking errors comparison for the two control techniques
purpose of clarity, we have chosen to plot the ZV input shaper response separately
(since the vertical axis scale are very distinct).
Amplitude (um)
Time (sec)
Amplitude (um)
Time (sec)
CONCLUSIONS
5.1 Conclusions
Due to advantages of PEA like small size, low cost, low power consumption,
ultra-high resolution, larger generated force etc, they are applied in precise
positioning applications, in general micro/nano positioning/manipulation
applications, in medical engineering applications etc. However, major setbacks for
piezoelectric actuators are hysteresis nonlinearity and residual vibration. These can
result in large positioning inaccuracies and may affect the system's stability. For
these reasons, control engineers investigate hysteresis and residual vibration
minimization techniques. In this work, literature review work on PEA modelling
techniques, hysteresis modelling techniques and hysteresis and vibration control
techniques has been conducted. Experimental processes were conducted in the
laboratory for the purpose of transient data collection, by subjecting the PEA system
to different input signals waveforms and frequencies. Image processing techniques
were used to process data acquired based on experiments to measure actuator's
displacement based by means of Euclidean distance measure from a reference point.
From the input and output data collected from experiments and FEA process, transfer
function models were developed and validity of the models were investigated.
Transient data showed that the actuator displacement was hysteretic in nature. To
develop the hysteresis model, Bouc-Wen model was used where Nelder Mead
simplex search algorithm was used for identifying optimum Bouc-Wen model
105
In order to extend this research work the following suggestion can contribute
in making the research work better. These include:
[14] Y. Li and Q. Xu, “Adaptive sliding mode control with perturbation estimation
and PID sliding surface for motion tracking of a piezo-driven
micromanipulator,” IEEE Trans. Control Syst. Technol., vol. 18, no. 4, pp.
798–810, July 2010.
[16] M. Špiller and Z. Hurák, “Hybrid charge control for stick–slip piezoelectric
actuators,” Mechatronics, vol. 21, no. 1, pp. 100–108, 2011.
[20] T. H. Saleh, “Active Vibration Control of a Smart Beam Using LQR , PID and
Fuzzy Logic Controllers,” M.S. thesis, Dept. Mech. Eng., University of
Basrah, Iraq, 2014.
108
[23] L. Gao, Q. Lu, F. Fei, L. Liu, Y. Liu, and J. Leng, “Active vibration control
based on piezoelectric smart composite,” Smart Mater. Struct., vol. 22, no. 12,
pp. 1-12, 2013.
[25] M.-J. Yang, C.-X. Li, G.-Y. Gu, and L.-M. Zhu, “A Modified Prandtl-
Ishlinskii Model for Rate-dependent Hysteresis Nonlinearity Using mth-power
Velocity Damping Mechanism,” Int. J. Adv. Robot. Syst., vol. 11, no. 163, pp.
1–10, 2014.
[27] J. Peng and X. Chen, ‘‘A Survey of Modeling and Control of Piezoelectric
Actuators,’’ Mod. Mech. Eng., vol. 03, no. 01, pp. 1–20, 2013.
[31] H. Q. Jerry, ‘‘Finite Element Analysis,’’ Lecture notes fall 2006, University of
Colorado Boulder, CO, USA, 2006.
[54] Z. Chi, M. Jia, and Q. Xu, “Fuzzy PID Feedback Control of Piezoelectric
Actuator with Feedforward Compensation,” Math. Probl. Eng., vol. 2014, pp.
1-14, 2014.
[55] Y. Xiao, Shunli and Li, “Modeling and High Dynamic Compensating the
Rate-Dependent Hysteresis of Piezoelectric Actuators via a Novel Modified
Inverse Preisach Model,” IEEE Trans. Control Syst. Technol., vol. 21, no. 5,
pp. 1549–1557, 2013.
111
[58] Y. Liu, J. Shan, U. Gabbert, and N. Qi, “Hysteresis and creep modeling and
compensation for a piezoelectric actuator using a fractional-order Maxwell
resistive capacitor approach,” Smart Mater. Struct., vol. 22, no. 11, pp. 1-12,
2013.
[68] S. K. S. Fan and E. Zahara, “A hybrid simplex search and particle swarm
optimization for unconstrained optimization,” Eur. J. Oper. Res., vol. 181, no.
2, pp. 527–548, 2007.
[70] S. Xiao and Y. Li, “Dynamic compensation and control for piezoelectric
actuators based on the inverse Bouc–Wen model,” Robot. Comput. Integr.
Manuf., vol. 30, no. 1, pp. 47–54, 2014.
[72] Y. Li, Z. Liu, and Y. Liu, “Hysteresis Modeling of Piezoelectric Actuators and
Feed-Forward Compensation Algorithm Research,” Int. J. Control Autom.,
vol. 8, no. 7, pp. 405–416, 2015.
[74] X. Zhang, K. Rall, and W. Papiernik, “Modified input shaping for circular
trajectory following,” In Proc. 7th WSEAS Int. Conf. Autom. Control,
Modeling Simul, 2005, pp. 271-276.
[75] Z. Mohamed and M. A. Ahmad, “Hybrid Input Shaping and Feedback Control
Schemes of a Flexible Robot Manipulator,” in Proceedings of the 17th World
Congress The International Federation of Automatic Control, 2008, pp. 6-11.
[76] H. Jiang, H. Ji, J. Qiu, and Y. Chen, “A modified prandtl-ishlinskii model for
modeling asymmetric hysteresis of piezoelectric actuators,” IEEE Trans.
Ultrason. Ferroelectr. Freq. Control, vol. 57, no. 5, pp. 1200–1210, 2010.
[80] C. Knospe, “PID control,” IEEE Control Syst. Mag., pp. 30–31, 2006.
[84] Quanser Inc., “Optimal lqr control,” Student workbook, Markham, Ontario,
Canada, 2014.
[85] K. Ogata, Modern Control Engineering 5th ed. New Jersy: Prentice Hall,
2010.
APPENDICES
APPENDIX A
List of Publications
4. Auwalu M. Abdullahi,
5. Auwal Shehu
116
APPENDIX B
Combined LQR, PID and pole placement vibration control simulation model
117
APPENDIX-C
void main() {
#define SignalPin 7;
int delayMs=500;
void setup() {
// One time configuration settings
pinMode(SignalPin,OUTPUT);
}
void loop() {
// put your main code here, to run repeatedly:
digitalWrite(SignalPin,LOW);
delay(delayMs);
digitalWrite(SignalPin,HIGH);
delay(delayMs);
}
}
118
APPENDIX-D
Table 0.1: Finite element model input and output data for transfer function
modelling using system identification technique.
Displacement
S/N Voltage (V) Displacement (m) S/N Voltage (V) (m)
1 0 0 22 78.75 2.27707E-07
2 3.75 1.08432E-08 23 82.50 2.38551E-07
3 7.500 2.16864E-08 24 86.25 2.49394E-07
4 11.25 3.25296E-08 25 90.00 2.60237E-07
5 15.00 4.33728E-08 26 93.75 2.7108E-07
6 18.75 5.4216E-08 27 97.50 2.81923E-07
7 22.50 6.50592E-08 28 101.25 2.92767E-07
8 26.25 7.59024E-08 29 105.00 3.0361E-07
9 30.00 8.67456E-08 30 108.75 3.14453E-07
10 33.75 9.75888E-08 31 112.50 3.25296E-07
11 37.50 1.08432E-07 32 116.25 3.36139E-07
12 41.25 1.19275E-07 33 120.00 3.46983E-07
13 45.00 1.30118E-07 34 123.75 3.57826E-07
14 48.75 1.40962E-07 35 127.50 3.68669E-07
15 52.50 1.51805E-07 36 131.25 3.79512E-07
16 56.25 1.62648E-07 37 135.00 3.90356E-07
17 60.00 1.73491E-07 38 138.75 4.01199E-07
18 63.75 1.84334E-07 39 142.50 4.12042E-07
19 67.50 1.95178E-07 40 146.25 4.22885E-07
20 71.25 2.06021E-07 41 150.00 4.33728E-07
21 75.00 2.16864E-07
119
APPENDIX E
Reading -1
Voltage Displacement Voltage Displacement
S/N (V) (um) S/N (V) (um)
1 0.063 0.000 41 0.063 0.000
2 0.063 0.000 42 0.063 0.000
3 63.050 6.543 43 63.050 6.666
4 63.050 6.543 44 63.050 6.666
5 0.063 0.000 45 0.063 0.123
6 0.063 0.000 46 0.063 0.123
7 63.050 6.543 47 63.050 6.543
8 63.050 6.543 48 63.050 6.543
9 0.063 0.000 49 0.063 0.123
10 0.063 0.000 50 0.063 0.123
11 63.050 6.666 51 63.050 6.666
12 63.050 6.666 52 63.050 6.666
13 0.063 0.123 53 0.063 0.000
14 0.063 0.123 54 0.063 0.000
15 63.050 6.543 55 63.050 6.666
16 63.050 6.543 56 63.050 6.666
17 0.063 0.000 57 0.063 0.123
18 0.063 0.000 58 0.063 0.123
19 63.050 6.420 59 63.050 6.666
20 63.050 6.420 60 63.050 6.666
21 0.063 0.000 61 0.063 0.123
22 0.063 0.000 62 0.063 0.123
23 63.050 6.543 63 63.050 6.666
24 63.050 6.543 64 63.050 6.666
25 0.063 0.000 65 0.063 0.000
120
Reading-2
Voltage Displacement Voltage Displacement
S/N (V) (um) S/N (V) (um)
81 0.063 0.000 121 0.063 0.000
82 0.063 0.000 122 0.063 0.000
83 63.050 6.790 123 63.050 6.543
84 63.050 6.790 124 63.050 6.543
85 0.063 0.123 125 0.063 0.000
86 0.063 0.123 126 0.063 0.000
87 63.050 6.666 127 63.050 6.666
88 63.050 6.666 128 63.050 6.666
89 0.063 0.123 129 0.063 0.123
90 0.063 0.123 130 0.063 0.123
91 63.050 6.666 131 63.050 6.666
92 63.050 6.666 132 63.050 6.666
93 0.063 0.123 133 0.063 0.123
94 0.063 0.123 134 0.063 0.123
95 63.050 6.790 135 63.050 6.666
96 63.050 6.790 136 63.050 6.666
97 0.063 0.123 137 0.063 0.000
98 0.063 0.123 138 0.063 0.000
99 63.050 6.666 139 63.050 6.666
100 63.050 6.666 140 63.050 6.666
121
Reading-3
Voltage Displacement Voltage Displacement
S/N (V) (um) S/N (V) (um)
161 0.063 0.123 201 0.063 0.247
162 0.063 0.123 202 0.063 0.247
163 63.050 6.790 203 63.050 6.790
164 63.050 6.790 204 63.050 6.790
165 0.063 0.123 205 0.063 0.247
166 0.063 0.123 206 0.063 0.247
167 63.050 6.666 207 63.050 6.666
168 63.050 6.666 208 63.050 6.666
169 0.063 0.000 209 0.063 0.123
170 0.063 0.000 210 0.063 0.123
171 63.050 6.790 211 63.050 6.790
172 63.050 6.790 212 63.050 6.79
173 0.063 0.247 213 0.063 0.247
174 0.063 0.247 214 0.063 0.247
175 63.050 6.666 215 63.050 6.790
122
Reading-4
Voltage Displacement Voltage Displacement
S/N (V) (um) S/N (V) (um)
241 0.063 0.247 281 0.063 0.247
242 0.063 0.247 282 0.063 0.247
243 63.050 6.666 283 63.050 6.790
244 63.050 6.666 284 63.050 6.790
245 0.063 0.123 285 0.063 0.247
246 0.063 0.123 286 0.063 0.247
247 63.050 6.790 287 63.050 6.790
248 63.050 6.790 288 63.050 6.790
249 0.063 0.123 289 0.063 0.247
250 0.063 0.123 290 0.063 0.247
123
Reading-5
Voltage Displacement Voltage Displacement
S/N (V) (um) S/N (V) (um)
321 0.063 0.171 361 0.063 0.344
322 0.063 0.171 362 0.063 0.344
323 63.050 6.900 363 63.050 6.900
324 63.050 6.900 364 63.050 6.900
325 0.063 0.344 365 0.063 0.344
124